Intelligent measuring and calculating system and method for curvature radius of plano-convex lens based on Newton ring interference

Abstract

The invention discloses a plano-convex lens curvature radius intelligent measuring and calculating system and method based on Newton ring interference, and relates to the technical field of optical measurement, and the system comprises a light path calibration unit which is used for enabling target light to enter a Newton ring device at a preset angle based on a calibration device; the automatic focusing unit is used for acquiring Newton ring images at different positions of the lens cone 8 based on the Newton ring device, screening the Newton ring images based on a pre-trained focusing scoring model to obtain a first image, obtaining a target position based on the first image, and adjusting the position of the lens cone 8 of the Newton ring device to the target position; the image acquisition unit is used for acquiring a second image based on the Newton ring device; the data measuring and calculating unit is used for classifying the second image based on a pre-trained image screening model to obtain an effective image with a central spot, and obtaining the curvature radius of a plano-convex lens based on the effective image, the problems that a traditional Newton ring measurement system is poor in adjustment precision, poor in reading accuracy, lack of automation degree and weak in external interference resistance are solved.

Description

Technical Field

[0001] This invention relates to the field of optical measurement technology, and more specifically, to an intelligent system and method for calculating the radius of curvature of a plano-convex lens based on Newton's rings interferometry. Background Technology

[0002] In existing techniques for Newton's rings experiments, conventional measuring devices have several limitations, severely impacting measurement accuracy and operational efficiency. Regarding instrument adjustment, when the human eye observes through the eyepiece, limited visual contrast can easily cause it to mistake a clear macroscopic area for the focal point, rather than the actual optimal focal plane. This results in insufficient focusing precision of the reading microscope's tube during focusing, requiring repeated adjustments of the mirror and tube to obtain a clearer image of Newton's rings. Furthermore, determining the center of the rings requires frequent movement of the Newton's rings element in conjunction with the rotation of the fine-tuning drum, a cumbersome process dependent on operator experience and exhibiting poor adjustment stability. Summary of the Invention

[0003] To address the issue of poor accuracy due to manual adjustment in traditional Newton's rings measurement systems, this application provides an intelligent calculation system for the radius of curvature of a plano-convex lens based on Newton's rings interferometry. The system includes:

[0004] Optical path calibration unit: used to direct the target light beam to the Newton's rings device at a preset angle based on the calibration device;

[0005] Autofocus unit: used to acquire Newton's rings images at different lens barrel positions based on the Newton's rings device, filter the Newton's rings images based on the focus scoring model to obtain a first image, obtain a target position based on the first image, and adjust the lens barrel position of the Newton's rings device to the target position;

[0006] Image acquisition unit: used to acquire several second images based on the Newton's rings device;

[0007] Data calculation unit: used to classify the second image based on a pre-trained image screening model to obtain an effective image with a central spot, and to obtain the radius of curvature of the plano-convex lens based on the effective image.

[0008] This system automatically acquires Newton's rings images under different lens barrels using a Newton's rings device. It then uses a focus scoring model to obtain and lock the optimal height of the lens barrel, determining the optimal focal plane for automatic and precise focusing. The Newton's rings device automatically adjusts the fine-tuning drum to capture images at the optimal focal plane. An image screening model eliminates blurry and invalid images, retaining only valid images that meet the requirements, resulting in smaller calculation errors. The center of the Newton's rings is determined by the Newton's rings in the valid images, thereby determining the radius of curvature. No manual adjustment or reading is required, making the operation simpler, more convenient, and the data more accurate.

[0009] Considering that the adjustment of the reflector in a telescope is usually done manually, and the mechanical structure lacks a precise 45° positioning device, it is difficult to accurately adjust the reflector to a position that forms an ideal 45° angle with the incident light source. This system uses a core board to drive a servo motor to rotate the reflector. An attitude sensor collects the reflector's attitude data in real time and provides closed-loop feedback to the core board. The reflector's tilt angle is displayed in real time through a client interface (such as Thonny) until the reflector tilt angle reaches 45°, ensuring that the illumination light is incident at a preset angle, thus achieving automatic and accurate adjustment of the reflector. At the same time, this system electrically connects both the servo motor and the attitude sensor to the core board, providing real-time feedback of the reflector data. This achieves a digital and automated interface, enabling real-time communication with computers or other data acquisition systems, eliminating the need for manual operation, improving efficiency and performance.

[0010] Furthermore, the calibration device includes a core board, a servo motor, and an attitude sensor. The servo motor and the attitude sensor are both connected to the reflector of the Newton's rings device, and the servo motor and the attitude sensor are both electrically connected to the core board.

[0011] Furthermore, the Newton's rings device includes a moving microscope, a first stepper motor, an industrial camera, and a second stepper motor. The industrial camera is fixed to the top of the microscope tube. The first stepper motor is used to adjust the coarse adjustment knob of the moving microscope, and the center of the first stepper motor and the coarse adjustment knob are at the same height. Both the first stepper motor and the industrial camera are connected to the client. The second stepper motor is used to adjust the fine adjustment drum of the moving microscope.

[0012] The first stepper motor carries the lens barrel and scans from bottom to top throughout the entire process. The client captures the Newton's rings image during the scanning process in real time, extracts the features of the Newton's rings image to generate a focus evaluation function, and plots it as a function value-height curve. The height corresponding to the peak of the curve is taken, and the first stepper motor is controlled to precisely fold back to that height and lock it, determining the optimal focal plane and achieving automatic and precise focusing. At the optimal focal plane, the second stepper motor automatically adjusts the fine adjustment drum to determine the center of the circle, eliminating the need for manual operation, making it simpler, faster, and more stable. Furthermore, the specific steps for obtaining the first image include:

[0013] Based on the focus scoring model, features of the Newton's rings image are extracted, and the features are fused to obtain a focus evaluation function. The first image is obtained based on the focus evaluation function. The features include Sobel gradient operator, FFT high-frequency energy, and Laplace variance.

[0014] The first calculation formula for the Sobel gradient operator is:

[0015] ;

[0016] in, This represents the Sobel gradient operator. This represents the height of the Newton's rings image. This represents the width of the Newton's Rings image. In the Newton's rings diagram, the first ring is represented by the second ring. Line number Column pixels in the horizontal direction gradient value, In the Newton's rings diagram, the first ring is represented by the second ring. Line number Column pixels in the vertical direction gradient value, and Both represent integers greater than or equal to 1;

[0017] The second formula for calculating the FFT high-frequency energy is:

[0018] ;

[0019] ;

[0020] in, Represents the high-frequency energy of FFT. Representing frequency point The energy at the location, Indicates the Newton's rings graph in coordinates grayscale value at that location Let represent the coordinates of the Newton's rings graph in the horizontal and vertical directions, respectively. Represents the imaginary unit. Represents pi (π). and Both represent discrete coordinates in the frequency domain. Indicates the high-frequency region. Indicates the number of frequency points within the high-frequency region;

[0021] The third formula for calculating the Laplace variance is:

[0022] ;

[0023] in, Indicates Laplace variance. Indicates the first [number]th ... Line number Pixel values ​​of column pixels, This represents the average value of all pixels within a local window.

[0024] The fourth calculation formula for the focus evaluation function is:

[0025] ;

[0026] in, This represents the focus evaluation function.

[0027] The focus scoring model is based on the characteristics of a clear Newton's ring image, such as significant gradient at the edge of the ring, rich high-frequency information, and dramatic changes in pixel grayscale. It constructs a focus evaluation function using the Sobel gradient operator, FFT high-frequency energy, and Laplace variance. The larger the function value, the higher the image clarity and the better the focus effect.

[0028] Furthermore, in accordance with the input order, the image screening model sequentially includes a feature extraction layer, an attention weight generation layer, a modality perception layer, a category reference vector library, a fusion layer, and a classification layer;

[0029] The feature extraction layer is used to convolve the second image to obtain a high-order feature map. The attention weight generation layer is used to process the high-order feature map to obtain attention-weighted features. The modality perception layer is used to compress the attention-weighted features to obtain a deep representation vector. The category reference vector library is used to obtain a weighted reference vector based on the deep representation vector. The fusion layer is used to fuse the deep representation vector and the weighted reference vector to obtain enhanced features. The classification layer is used to output a category prediction result based on the enhanced features.

[0030] Furthermore, according to the input order, the feature extraction layer includes several first layers, which sequentially include a convolutional layer, a batch normalization layer, a ReLU activation function, a max pooling layer, and a regularization layer; the attention weight generation layer sequentially includes a fully connected layer, the ReLU activation function, the fully connected layer, and a Sigmoid activation function; the modality perception layer sequentially includes the batch normalization layer, the fully connected layer, the regularization layer, and the fully connected layer.

[0031] Furthermore, considering that existing equipment relies entirely on manual visual interpretation during the observation and reading stages, the limited resolution of the human eye, coupled with visual fatigue from prolonged observation, often leads to errors in sequential interpretation and significant reading deviations. This results in large errors in the final calculated radius of curvature, failing to meet the high-precision measurement requirements of modern metrology. This system uses an industrial camera to capture images at equal angular intervals at the optimal focal plane. An image sieving model eliminates blurry and invalid images, retaining only valid images that meet the requirements. The valid images undergo global polar coordinate transformation to calculate the 360° radial grayscale gradient energy. The point corresponding to the maximum gradient energy is determined as the center coordinate of the circle. Using the center as the origin, a radial light intensity distribution array is obtained. The peak value of this array is found to obtain the dark ring pixel radius sequence. Concentric circles are drawn using the center coordinates and the dark ring pixel radius sequence to obtain the dark ring. A reference image of the micrometer is captured at the optimal focal plane, and a line is drawn on it to obtain a reference line. The pixel distance of the reference line is compared with the actual distance of the micrometer to obtain the conversion coefficient. Then, the dark ring pixel radius sequence is converted into the actual diameter according to the conversion coefficient, and the curvature radius of the plano-convex lens is calculated according to the curvature radius calculation formula. This achieves full automation from acquisition to calculation, eliminating the need for manual reading and operation, reducing data errors, and visualizing the calculation results.

[0032] Furthermore, the specific steps for obtaining the radius of curvature of the plano-convex lens based on the effective image include:

[0033] Several candidate points are selected within the search area to obtain candidate circle centers. A local polar coordinate system with the candidate circle center as the origin is constructed at each candidate circle center. Circumferential pixels are obtained within a preset radius range according to different radii and different angles. The average gray value of the circumferential pixels on the same radius is obtained, and the radius-light intensity distribution sequence corresponding to the candidate circle center is obtained.

[0034] Discretely differentiate the radius-intensity distribution sequence along the radial direction, accumulate the absolute values ​​of the intensity gradients at different radii to obtain the radial gradient energy of the candidate circle center; obtain the candidate circle center corresponding to the maximum value of the radial gradient energy to obtain the target circle center;

[0035] A first sequence is obtained based on the peak value of the radius-light intensity distribution sequence, a first curve is obtained based on the first sequence, and a number of pixel radii are obtained by acquiring the peak value of the first curve.

[0036] Within the image plane, concentric circles are drawn based on the target circle center and the pixel radius to obtain several dark rings. The dark rings are then sorted along the radial radius to obtain a dark ring sequence.

[0037] A reference image of the micrometer is obtained based on the target position. A reference line is obtained by drawing lines on the reference image. The pixel distance of the reference line is compared with the actual distance of the micrometer to obtain the conversion coefficient.

[0038] The radius of the dark ring sequence is converted based on the conversion coefficient to obtain the actual diameter of the dark ring sequence;

[0039] The radius of curvature is obtained based on the actual diameter and radius of curvature calculation formula.

[0040] The formula for calculating the radius of curvature is:

[0041] ;

[0042] in, and They represent the first The first dark ring and the first The actual diameter of the dark ring Indicates the radius of curvature. This indicates the wavelength of the sodium lamp. and Both represent the numbers of the dark rings, and Greater than .

[0043] Furthermore, the specific steps for obtaining the radius of curvature of the plano-convex lens based on the effective image also include:

[0044] A dark ring sequence is obtained based on the order distribution of the dark rings, and the radius sequence is obtained by calculating the dark ring radius of the dark ring sequence. Let the radius sequence If satisfied , If the condition is met, the dark cycle is determined to be an abnormal dark cycle and the dark cycle is deleted; if the condition is met... If the dark ring is determined to be a false dark ring, then the dark ring is deleted. Indicates the number of dark rings. They represent the first The and the first The radius of the dark ring Indicates the threshold for triggering an anomaly. Represents an integer greater than or equal to 1.

[0045] Based on the characteristic that the order and radius of Newton's rings are monotonically increasing and the difference between adjacent radii gradually increases, if the radius of a dark ring is smaller than the previous order, it is judged as noise interference and the abnormal dark ring is removed; if the difference between the radii of adjacent dark rings suddenly exceeds a preset multiple of the average difference of the preceding order, it is judged as a pseudo dark ring and removed.

[0046] Furthermore, the fifth formula for calculating the radial gradient energy is as follows:

[0047] ;

[0048] in, Indicated by Radial gradient energy when the candidate center is... This represents the x and y coordinates of the candidate circle center in the image coordinate system. Indicates the minimum radius value. Indicates the maximum radius value. This indicates the radius-intensity distribution sequence at the radius. The value at that location, This represents the radial distance between the candidate circle center and any sampling point. This represents the radial light intensity gradient.

[0049] This application also provides an intelligent method for calculating the radius of curvature of a plano-convex lens based on Newton's rings interference, the method comprising:

[0050] The calibration device directs the target light beam at a preset angle onto the Newton's rings apparatus.

[0051] The Newton's rings device acquires Newton's rings images at different lens barrel positions, filters the Newton's rings images based on the focus scoring model to obtain a first image, obtains a target position based on the first image, and adjusts the lens barrel position of the Newton's rings device to the target position.

[0052] Several second images were acquired based on the Newton's rings device;

[0053] The second image is classified based on a pre-trained image screening model to obtain an effective image with a central spot, and the radius of curvature of the plano-convex lens is obtained based on the effective image.

[0054] The principle and effect of this method are similar to those of this system, and therefore, no further details will be provided for this method.

[0055] One or more technical solutions provided by this invention have at least the following technical effects or advantages:

[0056] This system, through optical path calibration closed-loop control, autofocus, and dark ring fitting, significantly improves measurement accuracy, reduces the relative error in radius of curvature calculation, and meets metrological precision requirements. It can automate optical path calibration, autofocus, image acquisition, and processing with a single click, improving the efficiency of single-image output from acquisition to result. No manual calibration, reading, or calculation is required; it achieves automated and intelligent radius of curvature calculation, reducing the influence of external and human factors, minimizing errors, shortening manual operation time, and significantly improving experimental efficiency. Simultaneously, the system achieves digital and automated interfaces through calibration devices and Newton's rings devices, enabling real-time communication with computers or other data acquisition systems. All data recording, calculation, and error analysis are performed without manual intervention, improving efficiency, eliminating human error, and enhancing the reliability and reproducibility of experimental data processing. Attached Figure Description

[0057] The accompanying drawings, which are provided to further illustrate embodiments of the invention and constitute a part of this invention, are not intended to limit the scope of the invention.

[0058] Figure 1 This is a flowchart illustrating the intelligent calculation system for the radius of curvature of a plano-convex lens based on Newton's rings interference in this invention.

[0059] Figure 2 This is a schematic diagram of the Newton's rings apparatus;

[0060] Figure 3 This is a schematic diagram of the calibration device;

[0061] Among them, 1-core board, 2-servo motor, 3-attitude sensor, 4-first connector, 5-reflector, 6-moving microscope, 7-first stepper motor, 8-scope barrel, 9-industrial camera, 10-second connector, 11-coarse adjustment knob, 12-second stepper motor, 13-third connector, 14-fine adjustment drum. Detailed Implementation

[0062] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, where there is no conflict, the embodiments of the present invention and the features thereof can be combined with each other.

[0063] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0064] Example 1

[0065] refer to Figures 1-3 This embodiment provides an intelligent measurement system for the radius of curvature of a plano-convex lens based on Newton's rings interferometry. The system includes:

[0066] Optical path calibration unit: used to direct the target light beam to the Newton's rings device at a preset angle based on the calibration device;

[0067] Autofocus unit: used to acquire Newton's rings images at different lens barrel positions based on the Newton's rings device, filter the Newton's rings images based on the focus scoring model to obtain a first image, obtain a target position based on the first image, and adjust the lens barrel position of the Newton's rings device to the target position;

[0068] Image acquisition unit: used to acquire several second images based on the Newton's rings device;

[0069] Data calculation unit: used to classify the second image based on a pre-trained image screening model to obtain an effective image with a central spot, and to obtain the radius of curvature of the plano-convex lens based on the effective image.

[0070] In this embodiment, Newton's rings consist of a series of concentric rings with alternating light and dark colors. All these rings revolve around a common center, which is the central spot, corresponding to the position where the air film thickness is minimum (approaching zero).

[0071] The calibration device includes a core board 1 (such as an F103C8T6 STM32 core board 1), a servo motor 2 (such as an XH-8020 servo motor 2), and an attitude sensor 3 (such as an MPU9250 attitude sensor 3). Both the servo motor 2 and the attitude sensor 3 are connected to the reflector 55 of the Newton's rings device, and both are electrically connected to the core board 1. For example, the attitude sensor 3 is fixed to the back of the reflector 5 bracket via a 3D-printed mounting bracket. Its XYZ angles are read in real-time by the core board 1 and displayed at a refresh rate of 10Hz in the Thonny IDE serial monitor. The servo motor 2 is connected to the bracket via a first connector 4 (such as a 3D-printed connector) to achieve 0-90° closed-loop control.

[0072] The Newton's rings device includes a moving microscope 6, a first stepper motor 7 (such as an MS42DC stepper motor), an industrial camera 9 (such as a DW1600 industrial camera 9), and a second stepper motor 12 (such as an MS42DC stepper motor). The industrial camera 9 is fixed to the top of the microscope tube 8 of the moving microscope 6 by a 3D-printed clamp (25mm inner diameter, 5mm wall thickness) to ensure that the coaxiality between the camera optical axis and the microscope optical axis is ≤0.05mm. The first stepper motor 7 is used to adjust the coarse adjustment of the moving microscope 6. The first stepper motor 7 is connected to the coarse adjustment knob of the moving microscope 6 via a second connector 10 (such as a 3D printed connector), and the center of the first stepper motor 7 and the coarse adjustment knob 11 are at the same height. The first stepper motor 7 and the industrial camera 9 are both connected to the client for data. The second stepper motor 12 is used to adjust the fine adjustment drum 14 of the moving microscope 6. The fine adjustment drum 14 can be connected to the second stepper motor 12 via a third connector 13 (such as a 3D printed connector with a radius of 30mm).

[0073] In this embodiment, the 3D printed card holder, 3D printed clamp, first connector 4, second connector 10, third connector 13 and 3D printed connector are all used to connect components. Their size, shape, structure and material are determined by the two components that need to be connected, and they can be made using existing cutting technology, manufacturing technology, etc.

[0074] The autofocusing process is as follows: the first stepper motor 7 drives the lens barrel 8 to rotate from top to bottom to the highest point. Each rotation is used by the industrial camera 9 to capture one image. The client calculates the function value of the focus evaluation function for each image through the focus scoring model and plots the E-height curve. The height corresponding to the peak of the curve is selected as the optimal focal plane, and the first stepper motor 7 is controlled to reverse in a closed loop to that height and lock it.

[0075] Specifically, the process can be as follows: First, adjust the lens barrel 8 to the lowest point. Then, set the first stepper motor 7 to rotate at a speed of 10 rad / s, with a 5.4° step, a 2-second interval between each rotation, and a total of 40 rotations. At the same time, control the industrial camera 9 to acquire images every 2 seconds to achieve bottom-up scanning. The data is transmitted back to the client via USB cable, and a line graph of the E-angle is plotted. By finding the global maximum value, the system automatically controls the motor to reverse to the highest point.

[0076] The image acquisition steps are as follows: The client controls the second stepper motor 12 to drive the fine adjustment drum to rotate, which in turn moves the lens barrel 8 horizontally left and right. The industrial camera 9 simultaneously captures images of Newton's rings at the same angle to obtain the second image. Specifically, the second stepper motor 12 is driven by a Python program to move the lens barrel 8 at a speed of 10 rad / s, with a 1-second interval between each rotation, for a total of 60 rotations. At the same time, the industrial camera 9 is controlled to acquire images once every 1 second to obtain the second image.

[0077] At the same time, based on the image screening model, complete and valid images with a large number of rings are selected, while invalid images that are blurry or have no center are removed.

[0078] The specific steps for obtaining the first image include:

[0079] The Newton's rings image is extracted based on the focus scoring model (such as a deep learning model trained by SAM (Segment Anything Model) technology), the focus evaluation function is obtained by fusing the features, and the first image is obtained based on the focus evaluation function. The features include Sobel gradient operator, FFT high-frequency energy and Laplace variance.

[0080] The first calculation formula for the Sobel gradient operator is:

[0081] ;

[0082] in, This represents the Sobel gradient operator. This represents the height of the Newton's rings image. This represents the width of the Newton's Rings image. In the Newton's rings diagram, the first ring is represented by the second ring. Line number Column pixels in the horizontal direction gradient value, In the Newton's rings diagram, the first ring is represented by the second ring. Line number Column pixels in the vertical direction gradient value, and Both represent integers greater than or equal to 1;

[0083] The core principle of the Sobel gradient operator is to use two 3×3 convolution kernels to perform convolution operations on the image in the x and y directions, and calculate the gradient components of the pixels in these two directions. The larger the gradient, the more obvious the image edges and the more accurate the focus.

[0084] The second formula for calculating the FFT high-frequency energy is:

[0085] First, a two-dimensional discrete Fourier transform is performed on the Newton's rings image to obtain the spectrum:

[0086] ;

[0087] ;

[0088] in, Represents the high-frequency energy of FFT. Representing frequency point The energy at the location, Indicates the Newton's rings graph in coordinates grayscale value at that location Let represent the coordinates of the Newton's rings graph in the horizontal and vertical directions, respectively. Represents the imaginary unit. Represents pi (π). and Both represent discrete coordinates in the frequency domain. This indicates the high-frequency region, which is the high-frequency region in the frequency domain excluding the central low-frequency region. Indicates the number of frequency points within the high-frequency region;

[0089] The core principle of FFT high-frequency energy is to transform an image from the spatial domain to the frequency domain using Fast Fourier Transform (FFT) to obtain a two-dimensional spectrogram. The square of the amplitude at each point in the spectrogram represents the energy of that frequency component (according to Passevar's theorem, the total energy in the spatial domain is equal to the total energy in the frequency domain). After the transformation, the high-frequency energy corresponds to the region far from the center in the spectrogram, and its magnitude reflects the strength of high-frequency information such as edges and details in the image.

[0090] The third formula for calculating the Laplace variance is:

[0091] ;

[0092] in, Indicates Laplace variance. Indicates the first [number]th ... Line number Pixel values ​​of column pixels, This represents the average value of all pixels within a local window.

[0093] The core principle of Laplace variance is to detect the degree of grayscale change (edges and details) in an image and quantify the strength of this change with variance values, thereby determining the image's sharpness.

[0094] The fourth calculation formula for the focus evaluation function is:

[0095] ;

[0096] in, This represents the focus evaluation function.

[0097] The image screening model, in the order of input, includes a feature extraction layer, an attention weight generation layer, a modality perception layer, a category reference vector library, a fusion layer, and a classification layer.

[0098] The feature extraction layer is used to process the second image through multi-scale convolution to obtain a high-order feature map. The attention weight generation layer is used to perform adaptive gating on the high-order feature map to obtain attention-weighted features and realize feature recalibration. The modality perception layer is used to compress the attention-weighted features into a fixed dimension to obtain a depth representation vector. The category reference vector library is used to obtain a weighted reference vector based on the depth representation vector. The fusion layer is used to fuse the depth representation vector and the weighted reference vector element by element to obtain enhanced features. The classification layer is used to use a linear classifier to identify the enhanced features and output the category prediction result.

[0099] The category reference vector library is a learnable category reference vector library with a dimension of number of categories × 128. It is consistent with the output of the modality perception layer and is stored in the GPU memory as learnable parameters. It is used to generate weighted reference vectors in a class-related manner during the forward propagation stage.

[0100] In this configuration, the feature extraction layer comprises several first layers in the order of input. The first layers sequentially include a convolutional layer (Conv), a batch normalization layer (BN), a ReLU activation function, a max-pooling layer, and a dropout regularization layer. The attention weight generation layer sequentially includes fully connected layers (FC), the ReLU activation function, the fully connected layer, and a sigmoid activation function. The modality awareness layer sequentially includes the batch normalization layer, the fully connected layer, the regularization layer, and the fully connected layer.

[0101] The standard Newton's rings image is a series of concentric rings with alternating light and dark areas.

[0102] The specific steps for obtaining the radius of curvature of the plano-convex lens based on the effective image include:

[0103] Several candidate points are selected within the search area to obtain candidate circle centers. A local polar coordinate system with the candidate circle center as the origin is constructed at each candidate circle center. Circumferential pixels are obtained within a preset radius according to different radii and different angles. The average gray value of the circumferential pixels on the same radius (i.e., the arithmetic mean of the gray values ​​of all pixels) is obtained, and the radius-light intensity distribution sequence corresponding to the candidate circle center is obtained.

[0104] Discretely differentiate the radius-intensity distribution sequence along the radial direction, accumulate the absolute values ​​of the intensity gradients at different radii to obtain the radial gradient energy of the candidate circle center; obtain the candidate circle center corresponding to the maximum value of the radial gradient energy to obtain the target circle center;

[0105] For example, within the entire image or a user-specified search area, candidate circle centers are traversed in steps of 1-2 pixels. Based on the candidate circle centers and a preset radius range, circumferential pixels are obtained. For each candidate circle center, within the radius range... , Within the circle, circumferential pixels are sampled with a step size of 1° to obtain the average gray value of the circumferential pixels and obtain a one-dimensional radial radius-light intensity distribution sequence corresponding to the candidate circle center. The radius-light intensity distribution sequence is discretized along the radial direction (i.e., differential discretization). First, the radius-light intensity distribution sequence is differentiated, and then the result is discretized. The discretization process can be implemented by difference operation. By calculating the difference between adjacent data points, a light intensity difference sequence can be obtained. The absolute values ​​of the light intensity gradient (i.e., light intensity difference) at different radii are accumulated to obtain several radial gradient energies.

[0106] The fifth formula for calculating the radial gradient energy is as follows:

[0107] ;

[0108] in, Indicated by Radial gradient energy when the candidate center is... This represents the x and y coordinates of the candidate circle center in the image coordinate system. This represents the minimum radius value, i.e., the smallest radius used in the calculation. It is generally used to avoid areas with overexposure in the center or where the banding is not obvious. This represents the maximum radius value, i.e., the largest radius used in the calculation, which is jointly limited by the image size and the effective fringe range. This indicates the radius-intensity distribution sequence at the radius. The value at that point, that is, at a radius of At that time, with The average light intensity of the pixel grayscale obtained by uniformly sampling along the circumference of the center. This represents the radial distance between the candidate circle center and any sampling point, i.e., the radius distance from the candidate circle center to the sampling point along any direction. In actual calculations, [ , The interval is discretely selected with a fixed step size (e.g., 1 pixel). This represents the radial light intensity gradient, which is the first derivative of the light intensity with respect to the radius in the radial direction. In actual implementation, it is approximated by the difference in light intensity between adjacent radii.

[0109] A first sequence is obtained based on the peak value of the radius-light intensity distribution sequence, a first curve is obtained based on the first sequence, and a number of pixel radii are obtained by obtaining the peak value of the first curve; for the one-dimensional radial light intensity distribution sequence obtained above, its peak value is obtained, which is the pixel radius of the dark ring.

[0110] Within the image plane, concentric circles are drawn based on the target circle center and the pixel radius to achieve the fitting of dark rings, thereby obtaining several dark rings. The dark rings are sorted along the radial radius and numbered sequentially to obtain a dark ring sequence.

[0111] A reference image of the micrometer is obtained based on the target position. A reference line is obtained by drawing lines on the reference image. The pixel distance of the reference line is compared with the actual distance of the micrometer to obtain the conversion coefficient.

[0112] For example, after autofocusing, a micrometer image is taken at the optimal focus position using the device. The micrometer is similar to a ruler with very small precision. By drawing a line on the image on the client side and comparing the pixel distance of this line with the distance on the micrometer, the conversion coefficient between pixels and actual distance can be obtained.

[0113] The radius of the dark ring sequence is converted based on the conversion coefficient to obtain the actual diameter of the dark ring sequence; the radius of the dark ring is converted based on the conversion coefficient to obtain the actual radius, and only the actual diameter is obtained from this conversion coefficient.

[0114] The radius of curvature is obtained based on the actual diameter and radius of curvature calculation formula.

[0115] The formula for calculating the radius of curvature is:

[0116] ;

[0117] in, and They represent the first The first dark ring and the first The actual diameter of the dark ring Indicates the radius of curvature. This indicates the wavelength of the sodium lamp. and Both represent the numbers of the dark rings, and Greater than .

[0118] The specific steps for obtaining the radius of curvature of the plano-convex lens based on the effective image further include:

[0119] A dark ring sequence is obtained based on the order distribution of the dark rings, and the radius sequence is obtained by calculating the dark ring radius of the dark ring sequence. Let the radius sequence If satisfied , If the condition is met, the dark cycle is determined to be an abnormal dark cycle and the dark cycle is deleted; if the condition is met... If the dark ring is determined to be a false dark ring, then the dark ring is deleted. Indicates the number of dark rings. They represent the first The and the first The radius of the dark ring Indicates the threshold for triggering an anomaly. Represents an integer greater than or equal to 1.

[0120] in, Represents the current number Ring and the first The spacing between the dark rings between the rings. Represents from the first dark ring to the second. The average of all adjacent spacings that were considered normal between the dark rings. The abnormal trigger threshold is defined as how many times larger than the average value is considered an abnormal dark loop. In this embodiment, it is preferably 1.5, for the following reasons:

[0121] The spacing between normally adjacent dark rings decreases slowly, and the ratio of normal adjacent spacing is:

[0122] ;

[0123] Indicates the first The variation in the spacing between adjacent dark rings, Indicates the first The variation in the spacing between adjacent dark rings, that is, the spacing between adjacent rings is at most 20% to 40% smaller than the previous level, and will never suddenly become 1.5 times or more of the previous average.

[0124] And two scenarios where false dark rings may occur:

[0125] 1. A link was missing, namely... and The distance between them was taken as the spacing between adjacent pairs:

[0126] At this point, the following will occur:

[0127] ;

[0128] ;

[0129] This is 1.8 to 1.95 times the normal spacing.

[0130] 2. Mistaking a distant dark ring for a new ring:

[0131] At this point, the spacing is often much larger than the normal spacing, that is, twice or more than the normal spacing. Furthermore, due to experimental noise and other factors, H cannot be chosen too small, otherwise normal dark rings will be mistakenly deleted. Therefore, in this embodiment… A value of 1.5 is more appropriate.

[0132] Example 2

[0133] Based on Embodiment 1, in this embodiment, the system further includes a processing unit, which is specifically used for:

[0134] The effective image is preprocessed, and the radius of curvature is obtained based on the preprocessed effective image.

[0135] The preprocessing methods include, in order: grayscale conversion, contrast enhancement and noise suppression, morphological manipulation and skeletalization.

[0136] Grayscale conversion and contrast enhancement: Based on the core feature of Newton's rings alternating light and dark, the color image is converted to a grayscale image using OpenCV in Python, preserving the light and dark features of Newton's rings. Then, the CLAHE algorithm model is used to enhance the contrast of the image. The principle is to divide the image into 8×8 blocks, perform histogram equalization on the blocks, set an upper limit on the number of grayscale pixels, crop the excess pixels and distribute them evenly to other grayscale levels, and then eliminate the splicing gaps through interpolation to ensure the overall visual continuity of the final image.

[0137] Noise suppression includes Gaussian blur and bilateral filtering. Gaussian blur is used as an auxiliary denoising technique after contrast enhancement. Its principle is to calculate a weight matrix based on a two-dimensional Gaussian function, then perform a weighted average on all pixels within the window, and assign the result back to the center pixel. Because Gaussian weights decay rapidly with distance, neighboring pixels contribute more and distant pixels contribute less. This spatial correlation averaging can both pull isolated high-frequency noise points towards the neighborhood grayscale to suppress noise and partially preserve low-frequency contour and gradient information. Its basic formula is shown below:

[0138]

[0139] in, express, Represents the coordinates of any pixel within the domain. Indicates standard deviation, and It is the center of the Gaussian distribution.

[0140] Bilateral filtering is used for preprocessing before skeletonization. It multiplies spatial Gaussian weighted by a gray-level difference Gaussian weight, and performs weighted averaging only among spatially adjacent pixels with similar brightness, so as to smooth the internal region while preserving or even sharpening edge details.

[0141] Morphological operations include morphological opening and closing operations. Closing operation: using a 7×7 elliptical structure operation to fill small dark gaps in the main area of ​​Newton's rings and repair contour breaks; Opening operation: using a 5×5 elliptical structure operation to eliminate bright noise smaller than the structural element in the image and preserve the main contour of Newton's rings.

[0142] The opening operation works by using a structuring element to eliminate bright noise smaller than the structuring element in the image and restore the main subject area; the closing operation fills in small dark gaps within the main subject area and repairs broken contours. Combining the two allows for targeted noise removal and gap repair while preserving the main contour of the Newton's rings.

[0143] For the skeletonization process, the grayscale image is first converted into a binary image using an adaptive Gaussian threshold to highlight the edge of the ring. Then, the binarized ring edge is refined into a skeleton of single-pixel width by cyclically performing erosion-dilation-difference operations, which greatly reduces redundant pixels. Finally, a 3×3 cross kernel is used to perform a dilation operation on the skeleton image to restore the continuity of the ring band, which is convenient for subsequent calculation of the center and radius.

[0144] Employing a multi-level image processing algorithm combining CLHAE, Gaussian blur, bilateral filtering, and morphological operations, it possesses adaptive repair capabilities for defects such as uneven illumination, sensor noise, and ring breakage.

[0145] Example 3

[0146] Based on the above embodiments, this embodiment also provides an intelligent method for calculating the radius of curvature of a plano-convex lens based on Newton's rings interference, the method comprising:

[0147] The calibration device directs the target light beam at a preset angle onto the Newton's rings apparatus.

[0148] Based on the acquisition of Newton's rings images at different positions of the lens barrel 8 using the Newton's rings device, the first image is obtained by filtering the Newton's rings images based on the focus scoring model, the target position is obtained based on the first image, and the position of the lens barrel 8 of the Newton's rings device is adjusted to the target position.

[0149] Several second images were acquired based on the Newton's rings device;

[0150] The second image is classified based on a pre-trained image screening model to obtain an effective image with a central spot, and the radius of curvature of the plano-convex lens is obtained based on the effective image.

[0151] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0152] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An intelligent system for calculating the radius of curvature of a plano-convex lens based on Newton's rings interference, characterized in that, The system includes: Optical path calibration unit: used to direct the target light beam to the Newton's rings device at a preset angle based on the calibration device; Autofocus unit: used to acquire Newton's rings images at different lens barrel positions based on the Newton's rings device, filter the Newton's rings images based on the focus scoring model to obtain a first image, obtain a target position based on the first image, and adjust the lens barrel position of the Newton's rings device to the target position; Image acquisition unit: used to acquire several second images based on the Newton's rings device; Data calculation unit: used to classify the second image based on a pre-trained image screening model to obtain an effective image with a central spot, and to obtain the radius of curvature of the plano-convex lens based on the effective image.

2. The intelligent calculation system for the radius of curvature of a plano-convex lens based on Newton's rings interference as described in claim 1, characterized in that, The calibration device includes a core board, a servo motor, and an attitude sensor. The servo motor and the attitude sensor are both connected to the reflector of the Newton's rings device, and the servo motor and the attitude sensor are both electrically connected to the core board.

3. The intelligent calculation system for the radius of curvature of a plano-convex lens based on Newton's rings interference as described in claim 1, characterized in that, The Newton's rings device includes a moving microscope, a first stepper motor, an industrial camera, and a second stepper motor. The industrial camera is fixed to the top of the microscope tube. The first stepper motor is used to adjust the coarse adjustment knob of the moving microscope, and the center of the first stepper motor and the coarse adjustment knob are at the same height. Both the first stepper motor and the industrial camera are connected to the client. The second stepper motor is used to adjust the fine adjustment drum of the moving microscope.

4. The intelligent calculation system for the radius of curvature of a plano-convex lens based on Newton's rings interference as described in claim 1, characterized in that, The specific steps to obtain the first image include: Based on the focus scoring model, features of the Newton's rings image are extracted, and the features are fused to obtain a focus evaluation function. The first image is obtained based on the focus evaluation function. The features include Sobel gradient operator, FFT high-frequency energy, and Laplace variance. The first calculation formula for the Sobel gradient operator is: ； in, This represents the Sobel gradient operator. This indicates the height of the Newton's rings image. This represents the width of the Newton's Rings image. In the Newton's rings diagram, the first ring is represented by the second ring. Line number Column pixels in the horizontal direction gradient value, In the Newton's rings diagram, the first ring is represented by the second ring. Line number Column pixels in the vertical direction gradient value, and Both represent integers greater than or equal to 1; The second formula for calculating the FFT high-frequency energy is: ； ； in, Represents the high-frequency energy of FFT. Representing frequency point The energy at the location, Indicates the Newton's rings graph in coordinates grayscale value at that location Let represent the coordinates of the Newton's rings graph in the horizontal and vertical directions, respectively. Represents the imaginary unit. Represents pi (π). and Both represent discrete coordinates in the frequency domain. Indicates the high-frequency region. Indicates the number of frequency points within the high-frequency region; The third formula for calculating the Laplace variance is: ； in, This represents the Laplace variance. Indicates the first [number]th ... Line number Pixel values ​​of column pixels, This represents the average value of all pixels within a local window. The fourth calculation formula for the focus evaluation function is: ； in, This represents the focus evaluation function.

5. The intelligent calculation system for the radius of curvature of a plano-convex lens based on Newton's rings interference as described in claim 1, characterized in that, According to the input order, the image screening model includes a feature extraction layer, an attention weight generation layer, a modality perception layer, a category reference vector library, a fusion layer, and a classification layer. The feature extraction layer is used to convolve the second image to obtain a high-order feature map. The attention weight generation layer is used to process the high-order feature map to obtain attention-weighted features. The modality perception layer is used to compress the attention-weighted features to obtain a deep representation vector. The category reference vector library is used to obtain a weighted reference vector based on the deep representation vector. The fusion layer is used to fuse the deep representation vector and the weighted reference vector to obtain enhanced features. The classification layer is used to output a category prediction result based on the enhanced features.

6. The intelligent calculation system for the radius of curvature of a plano-convex lens based on Newton's rings interference as described in claim 5, characterized in that, According to the input order, the feature extraction layer includes several first layers, which in turn include a convolutional layer, a batch normalization layer, a ReLU activation function, a max pooling layer, and a regularization layer; the attention weight generation layer in turn includes a fully connected layer, the ReLU activation function, the fully connected layer, and a Sigmoid activation function; the modality perception layer in turn includes the batch normalization layer, the fully connected layer, the regularization layer, and the fully connected layer.

7. The intelligent calculation system for the radius of curvature of a plano-convex lens based on Newton's rings interference as described in claim 1, characterized in that, The specific steps for obtaining the radius of curvature of the plano-convex lens based on the effective image include: Several candidate points are selected within the search area to obtain candidate circle centers. A local polar coordinate system with the candidate circle center as the origin is constructed at each candidate circle center. Circumferential pixels are obtained within a preset radius range according to different radii and different angles. The average gray value of the circumferential pixels on the same radius is obtained, and the radius-light intensity distribution sequence corresponding to the candidate circle center is obtained. Discretely differentiate the radius-intensity distribution sequence along the radial direction, accumulate the absolute values ​​of the intensity gradients at different radii to obtain the radial gradient energy of the candidate circle center; obtain the candidate circle center corresponding to the maximum value of the radial gradient energy to obtain the target circle center; A first sequence is obtained based on the peak value of the radius-light intensity distribution sequence, a first curve is obtained based on the first sequence, and a number of pixel radii are obtained by acquiring the peak value of the first curve. Within the image plane, concentric circles are drawn based on the target circle center and the pixel radius to obtain several dark rings. The dark rings are then sorted along the radial radius to obtain a dark ring sequence. A reference image of the micrometer is obtained based on the target position. A reference line is obtained by drawing lines on the reference image. The pixel distance of the reference line is compared with the actual distance of the micrometer to obtain the conversion coefficient. The radius of the dark ring sequence is converted based on the conversion coefficient to obtain the actual diameter of the dark ring sequence; The radius of curvature is obtained based on the actual diameter and radius of curvature calculation formula. The formula for calculating the radius of curvature is: ； in, and They represent the first The first dark ring and the first The actual diameter of the dark ring Indicates the radius of curvature. This indicates the wavelength of the sodium lamp. and Both represent the numbers of the dark rings, and Greater than .

8. The intelligent calculation system for the radius of curvature of a plano-convex lens based on Newton's rings interference according to claim 7, characterized in that, The specific steps for obtaining the radius of curvature of the plano-convex lens based on the effective image also include: A dark ring sequence is obtained based on the order distribution of the dark rings, and the radius sequence is obtained by calculating the dark ring radius of the dark ring sequence. Let the radius sequence If satisfied , If the condition is met, the dark cycle is determined to be an abnormal dark cycle and the dark cycle is deleted; if the condition is met... If the dark ring is determined to be a false dark ring, then the dark ring is deleted. Indicates the number of dark rings. They represent the first The and the first The radius of the dark ring Indicates the threshold for triggering an anomaly. Represents an integer greater than or equal to 1.

9. The intelligent calculation system for the radius of curvature of a plano-convex lens based on Newton's rings interference as described in claim 8, characterized in that, The fifth formula for calculating the radial gradient energy is: ； in, Indicated by Radial gradient energy when the candidate center is... This represents the x and y coordinates of the candidate circle center in the image coordinate system. Indicates the minimum radius value. Indicates the maximum radius value. This indicates the radius-intensity distribution sequence at the radius. The value at that location, This represents the radial distance between the candidate circle center and any sampling point. This represents the radial light intensity gradient.

10. A method for intelligently calculating the radius of curvature of a plano-convex lens based on Newton's rings interference, characterized in that, The method includes: The calibration device directs the target light beam at a preset angle onto the Newton's rings apparatus. The Newton's rings device acquires Newton's rings images at different lens barrel positions, filters the Newton's rings images based on the focus scoring model to obtain a first image, obtains a target position based on the first image, and adjusts the lens barrel position of the Newton's rings device to the target position. Several second images were acquired based on the Newton's rings device; The second image is classified based on a pre-trained image screening model to obtain an effective image with a central spot, and the radius of curvature of the plano-convex lens is obtained based on the effective image.

Images